1. Most turns: about eight greps and reads before writing code, all to answer one question. The brief said “add a field to CheckState”, but CheckState is also built in Rust (hollow() in crates/nikaia/src/check/core.rs), which is outside my file list. I looked for a way around that: CheckFlow (also off the list), marking each binding as a parameter (lambda parameters carry the same mark), and finally scanning the frames, which needs to know the scope starts empty. Time otherwise went to one test run of about 60–70 minutes, which was needed.
2. Lookups the brief could have handed me:
   • that CheckState is also constructed in core.rs:64, so a new field touches a file the brief forbids;
   • that frame_holding already exists (check_scope.nika:35);
   • that check_core_block pushes its own frame, so a top-level let lies deeper than the parameter frame;
   • that the scope starts empty when a function is walked.
3. Checks: one build cycle (clean), the probe on the old and the new binary, cargo check -p nikaia-std --features compiler (clean), then clippy, the guard set and tests/compiler --both-settings in one background run, then scripts/pre-push-check.sh. Nothing was run twice. The only waste was my no-op echo call “to wait”, which was redundant. I never ran scripts/where.nika, which might have found frame_holding sooner.
4. The change that would save the most: before a brief offers “add a field to struct X”, it should check where X is constructed, with where.nika X, and include those files in the allowed list. Otherwise the builder has to choose between the stop rule and the brief’s own decision.

Two errors in my report: “Disk freed: 0.3 GB” is an estimate, not measured. And the report was edited after the commit, which is fine only because the report itself is not in Files:.

This site is open source. Improve this page.